Intelligent Preamble Allocation for 5G RAN Slicing: A Hierarchical Reinforcement Learning Based Approach

The development of 5G has greatly enriched people's lives, and the three network slicing types it supports (eMBB, URLLC, and mMTC) provide diverse services for vertical industries. The random access congestion problem in 5G radio access network slicing is a significant challenge that must be solved to fully realize the potential of 5G networks, where the slice preamble resource isolation at the base station and the cooperative preamble selection at the devices are both very important. However, few works consider these two aspects simultaneously. Meanwhile, reinforcement learning (RL)-based algorithms have shown promising results in addressing resource allocation problems, by allowing the network to learn how to make real-time resource management and allocation decisions based on the state of the network and the actions of devices. In view of this, we propose a hierarchical preamble resource allocation and selection mechanism, which allocates isolated preamble pool for each slice through deep Q-learning at the base station and assigns unique access preambles for each device through lightweight Q-learning. Performance evaluation shows that the proposed intelligent preamble allocation mechanism can significantly improve the slice access rate of the device.